Accelerated motion can be analyzed just fine from the perspective of an inertial frame (and technically a modern physicist would generally say that even if you use a non-inertial frame with a pseudo-gravitational field as discussed
here, this is still 'SR' if spacetime is not curved). Any smoothly-curve path in spacetime can be approximated as a polygonal path made up of a series of straight segments of constant velocity joined by instantaneous accelerations, and if you take the limit (in the calculus sense) as the segments become shorter and shorter (and the number of segments becomes greater and greater), this should approach perfect agreement with the original smooth path. If a given straight segment has a velocity v as seen in the inertial observer's frame, and the time between the beginning and end of the segment in the inertial observer's frame is dt, then the proper time along that segment should be \sqrt{1 - v^2/c^2} dt. So, if you have a smoothly-curved path where the velocity as a function of time in the inertial observer's frame is some function v(t), the proper time along this path between two moments t
0 and t
1 (which could be the moments of departing from and reuniting with a twin, for example) would be given by the integral \int_{t_0}^{t_1} \sqrt{1 - v(t)^2/c^2}\,dt (because in calculus an integral is just the limiting case of a sum in which the size of each segment dt becomes arbitrarily short)